CrowdPupil: A crowdsourced, pupil-center annotated image dataset

We present a dataset of pupil images and associated hand-annotated pupil centers, obtained through the method of crowd-sourcing. Acquisition of the points is explained and the dataset is presented. We present a comparison of two state-of-the-art pupil detection algorithms as a proposal towards public benchmarking of pupil detection algorithms. We invite the eye tracking community to test their own algorithms, share the results, and thereby advance the domain systematically. We present our plans for organizing a public pupil-detection challenge.